Identification of SDG Proteins in the D. catenatum Genome
To obtain all the members of SDG proteins in D. catenatum, we performed BLASTP search using known Arabidopsis and rice SDG proteins as queries against D. catenatum genome (INSDC: JSDN00000000.2). First, we checked the SDG genes of Arabidopsis and rice in the Superfamily 1.75 database (http://supfam.org/SUPERFAMILY/), we discovered 49 genes in Arabidopsis thaliana, corresponding to those reported in literature (Additional file 1) [10, 17]. On the other hand, 46 genes were identified in Oryza sativa (Additional file 1), three (Os01g65730/OsSET44, Os01g74500/OsSET45, Os06g03676/OsSET46) more than the 43 reported genes . Reciprocal BLAST was carried out to confirm that the hits from D. catenatum and its close relative Phalaenopsis equestris belong to the SDG family. Finally, we obtained 44 SDG genes in D. catenatum (Table 1 and Fig. 1) and 42 in P. equestris (Additional file 2), and they were named after their Arabidopsis homologs.
To characterize and classify the SDG family in D. catenatum, we used SDG proteins in the dicot model plant A. thaliana, the monocot model plant O. sativa, and its close relative P. equestris as references for phylogenetic analysis. The results showed that the 181 SDG proteins from the above four species could be clustered into eight classes (I~VII and M), mostly corresponding to the classification criteria in Arabidopsis . By contrast, ARABIDOPSIS TRITHORAX RELATED 3 (ATXR3) branch was previously classified into Class III , but it was separated from the other branches and near ATXR7-like proteins in Class IV in this study (Fig. 1). Considering their similar substrate specificities, we combined ATXR3 branch and the neighboring ATXR7-like proteins and categorized them under class M, as supported by the further phylogenetic analysis among Classes III, IV, and M (Fig. 4). Given the high limited reports on Class VI and VII, which feature potential functions for non-histone and histone methylation, we mainly focused on the roles of Classes I~V and M with well-investigated histone methylation specificity in this study.
Class I: E(z)-Like (H3K27me2/3)
Class I contains two E(z) homologs in each of monocot plants D. catenatum, P. equestris and rice, and three well-characterized homologs, namely, CURLY LEAF (CLF), SWINGER (SWN), and MEAEA (MEA), that represent three distinct clades in the dicot plant Arabidopsis (Fig. 2). The genes in this class contain 15~16 introns, which are extremely longer in the two orchid species compared with those of Arabidopsis and rice. This result suggests that overall intron length positively correlated with the corresponding genome size. A similar phenomenon related to intron/exon proportion was also observed in the members of the other classes as will be mentioned later. The three Arabidopsis E(z) proteins act as the catalytic subunits of the evolutionarily conserved Polycomb Repressive Complex 2 (PRC2), which is involved in the deposition of H3K27me3 repressive mark on the target gene locus . CLF (dominant H3K27me3 writer) and SWN act redundantly in vegetative and reproductive development, whereas MEA functions exclusively in suppression of central cell proliferation and endosperm development [45-47]. Rice E(z) homologs SDG711 and SDG718 participate in mediating accurate photoperiod control of flowering time . Clades I-1 (CLF-like) and I-2 (SWN-like) each contain one ortholog in the examined species, but Clade I-3 (MEA-like) is confined to Arabidopsis. Plant E(z)-like proteins generally harbor highly conserved domain organization at the C-terminal region, which includes a SANT domain, the cysteine rich CXC domain, and the signature SET domain, except for DcCLF and PeSWN, which lack the SANT domain, and DcSWN, which possesses an additional SANT domain at the N-terminus.
Class II: Ash-Like (H3K36me)
Class II can be further divided into five clades, each of which consists of a single member per plant species, except for two members of Clade II-1 in Arabidopsis and D. catenatum (Fig. 3). Clades II-1 to II-4 exist in all the examined species, but Clade II-5 is only found in rice and contains a single member, SDG707, with unknown function. Class II proteins generally share three conserved domains: an Associated with SET (AWS), SET, and PostSET domains [10, 49].
Clade II-1 (ASHH3-like) members are relatively shorter than their homologs in the four other clades. Arabidopsis ASHH3/SDG7 is required for proper timing in response to vernalization . SDG7 lacks detectable HKMTase activity , but rice ortholog SDG724/LVP possesses H3K36 methylation activity. The loss of SDG724 leads to late flowering . Notably, DcASHH3a/3b in D. catenatum lack AWS domain, different from PeASHH3 from its close relative P. equestris and ASHH3 orthologs in Arabidopsis and rice, and their functional divergence during speciation is interesting to investigate.
Clade II-2 (ASHR3-like) members are characterized by an additional PHD domain near the N-terminus, except for rice SDG736. ASHR3/SDG4 participates in regulating pollen tube growth and stamen development, and its overexpression leads to growth arrest and male sterility [53, 54]. ASHR3 harbors catalytic activities on H3K36me1 and possible H3K36me2, which is involved in regulating cell division competence in the root meristem .
Clade II-3 (ASHH1-like) members display uniform protein length and highly conserved AWS-SET-PostSET domain combination at the N-terminus. Arabidopsis ASHH1/SDG26 knockout leads to a late-flowering phenotype through decreasing H3K4me3 and H3K36me3 level at the SOC1 locus [56, 57]. Similarly, the knockdown of rice ortholog SDG708 causes a late-flowering phenotype and a genome-wide decrease in H3K36me1/2/3 levels during early growth stages . Predictably, D. catenatum DcASHH1 harbors a similar function.
Clade II-4 (ASHH2-like) proteins are considerably longer than the others, and characterized by an additional CW domain near to the N-terminal triple domain combination. Arabidopsis ASHH2/SDG8 acts as the major H3K36me2/3 writer [57, 59], and its knockout leads to pleiotropic phenotypes in vegetative and reproductive stage . Consistently, the knockdown of rice ortholog SDG725 causes wide-ranging defects, including dwarfism, erect leaves and small seeds . In the aspect of protein architecture, ASHH2 ortholog in D. catenatum or P. equestris is more like Arabidopsis SDG8 than rice SDG725.
Class III: Trx/ATX-Like (H3K4me2/3)
Class III consists of five members, which can be further divided into three clades in each examined plant species (Fig. 4). Class III proteins characterized tandem PHD domains in the middle region and SET–PostSET domain combination at the N-terminus. Moreover, several clades contain additional distinct domains, such as PWWP domain specific to Clade III-1/2, and FYRN–FYRC domain combination specific to Clade III-1.
Clade III-1 (ATX1-like) contains two members in Arabidopsis, and one in each of the three other species. In Arabidopsis, ATX1 and ATX2 paralogs exhibit similar domain architectures , but have distinct expression patterns in most cases and influence the expressions of largely nonoverlapping gene sets . For the shared targets, ATX1 and ATX2 account for the deposition of H3K4me3 and H3K4me2 marks, respectively . Different from Arabidopsis atx1 with early-flowering phenotype , rice ortholog mutant ostrx1/sdg723 exhibits late-flowering through decreased H3K4me3 levels at the central flowering time integrator Ehd1. OsTRX1 can rescue Arabidopsis atx1 phenotype [63-65], suggesting that ATX1-like proteins demonstrate conserved biochemical and molecular functions during evolution. However, ATX1-like proteins produce specific phenotypes in distinct species due to the differences in developmental context. Thus, DcATX1 and PeATX1 in orchid may play important roles in flowering time control.
Clade III-2 (ATX3-like) includes three members in each tested species. Arabidopsis ATX3/4/5 are clustered together and separated from the monocot orthologs. The orthologs from D. catenatum, and P. equestris are consistently clustered together, concordant with their close relationship. In Arabidopsis, ATX3/4/5 exhibit a common evolutionary origin, and function redundantly in genome-wide H3K4me2/3 profiles. Furthermore, atx3 atx4 atx5 triple mutant displays dwarfism and reduced fertility . In rice, ATX3-like proteins SDG721 and SDG705 function redundantly in modulating H3K4 methylation levels. The loss of both genes results in semi-dwarfism . Considering the dwarf phenotype of ATX3-like mutants in Arabidopsis and rice, the homologs in D. catenatum and P. equestris might be involved in regulating plant architecture.
Clade III-3 shows specificity toward the examined monocots and contains one copy per species. D. catenatum DcATX3d and P. equestris PeATX3d are characterized by an additional Jas domain at the C-terminus, in contrast with the rice ortholog OsSET37/SDG732. Further survey of this clade will be helpful for understanding the evolution of SDG family in monocots.
Class M: ATXR3/7 (H3K4me)
Class M comprises of two clades, namely, Clade M-1 (ATXR7-like) and M-2 (ATXR3-like). Each clade contains one copy per plant species (Fig. 4). ATXR7-like proteins usually lack extra domains, except for PeATXR7 with a C-terminal GYF domain. Arabidopsis ATXR7/SDG25 acts as the writer of H3K4 monomethylation (H3K4me1), and its knockout results in early flowering [59, 68]. ATXR3-like proteins also contain only one copy in each species, are characterized by the presence of DUF4339 domain in the middle region, except for OsSET27/SDG701. Arabidopsis ATXR3/SDG2 is the major H3K4me3 writer, whose depletion leads to pleiotropic development defects [28, 69, 70]. D. catenatum DcATXR3 and P. equestris PeATXR3 feature a more similar protein architecture to Arabidopsis ATXR3/SDG2 than rice SDG701. This finding suggests that ATXR3-like proteins in orchid may retain their ancestral role, whereas rice ortholog may functionally diverge, as attributed to the loss of specific domain and partial sequence.
Class IV: Su(var)-like (H3K27me1)
Class IV can be divided into two clades, Clade IV-I (ATXR5-like) and IV-II (ATXR6-like), which are characteristic of an N-terminal PHD domain in addition to the defined SET domain, except for PeATXR5 (Fig. 4). In Arabidopsis, ATXR5 and ATXR6 show largely overlapping functions, and the depletion of both results in global H3K27me1 reduction and heterochromatin decondensation [71, 72]. ATXR5/6 are involved in maintaining DNA replication  and repressing the expression of transposable element . The overexpression of either ATXR5 or ATXR6 causes male sterility . ATXR5 and ATXR6 probably perform separate roles because of ATXR5 with a dual localization in plastids and nucleus but ATXR6 solely in nucleus .
Class V: Suv-Like (H3K9me)
Class V contains 15 members in Arabidopsis, 14 in rice and D. catenatum, and 13 in P. equestris; These members can be further divided into two subclasses, SUVH and SUVR, which include Clades V-1 to V-3 and V-4 to V-6, respectively (Fig. 5). Class V proteins are usually characterized by PreSET–SET–PostSET or PreSET–SET domain combinations. SUVH proteins often contain another symbolic SET- and RING-ASSOCIATED (SRA) domain, whereas SUVR proteins in Clades V-4 and V-5 often include another WIYLD domain and tandem ZnF_C2H2 domains, respectively. SUVH genes usually lack introns, except for the members of SUVH4 branch and two members (PeSUVH45 and SDG727) of SUVH5 branch, whereas SUVR genes contain variable number of introns. In general, Class V members are responsible for methylation of histone H3 lysine 9 (H3K9me), in which H3K9 dimethylation (H3K9me2) is the critical mark for gene silencing and DNA methylation, and are involved in heterochromatin formation and reprogramming of gene expression .
In Clade V-1, the five members SUVH1/3/7/8/10 in Arabidopsis cluster together, and are distinct from the five homologs in rice and each of the two homologs in D. catenatum or P. equestris. This result indicates that duplication of these clade members occurred after divergence between dicots and monocots. However, the two orthologs in either D. catenatum or P. equestris respectively pair together, indicating that their gene duplication occurred before the split of Dendrobium and Phalaenopsis. Arabidopsis SUVH1/SDG32 performs a distinct anti-silencing function to promote the expression of DNA methylation-targeted genes. SUVH1 knockout causes no effect on H3K9me2 levels but reduces H3K4me3 levels . Furthermore, SUVH1 binds to highly methylated genomic loci targeted by RNA-directed DNA methylation (RdDM). However, rice SUVH1-like protein SDG728 retains its classical function to mediate H3K9 methylation and participate in retrotransposon repression . D. catenatum and P. equestris include two SUVH1-like proteins, far less than the five members in Arabidopsis and rice. Thus, the function of SUVH1 homologs and the evolutionary mechanisms in orchids require further investigation.
Clade V-2 comprises two members for each examined species, and these members lack PostSET domains, in contrast with those in clade V-1. In Arabidopsis, SUVH2 and SUVH9 as sister paralogs show overlapping functions in RdDM and heterochromatic gene silencing [79, 80]. SUVH2 overexpression leads to ectopic heterochromatization accompanied with significant developmental defects, such as extreme dwarfism [79, 81]. SUVH2 and SUVH9 may feature inactive histone methyltransferase activity [82, 83]. However, the simultaneous absences of SUVH2 and SUVH9 lead to a marked decrease in H3K9me2 levels in the RdDM loci [80, 84]. SUVH2 and SUVH9 can bind to methylated DNA and facilitate the recruitment of Pol V to RdDM loci [82, 84]. Considering the highly similar domain organization among SUVH2-like proteins in these examined species, their function should be evolutionarily conserved.
Clade V-3 proteins could be further divided into two branches (SUVH4 and SUVH5). SUVH4 branch possesses one member in each species, whereas SUVH5 branch contains two members in Arabidopsis and rice, and three members in D. catenatum or P. equestris. In Arabidopsis, SUVH4/KRYPTONITE (KYP), SUVH5, and SUVH6 as H3K9 methyltransferases are required to maintain DNA methylation [85-90]. SUVH4 as the predominant H3K9me1/2 writer . SUVH5 and SUVH6 as sister paralogs in SUVH5 branch exhibit HKMTase activities with locus-specific features [76, 86, 87]. In rice, SUVH4-like protein SDG714 mediates H3K9 methylation, participating in DNA methylation, transposition of transposable elements, and genome stability . Notably, PeSUVH4 protein in P. equestris is evidently short and lack SRA and PostSET domains, compared with SUVH4-like proteins in the three other species. The divergence of SUVH4 between orchid genus Dendrobium and phalaenopsis is worthy of investigation.
In Clade V-4, there are 3 members in Arabidopsis and P. equestris, 1 in rice, and 4 in D. catenatum, respectively. In Arabidopsis, SUVR2 mediates transcriptional silencing in both RdDM-dependent and -independent manners . SUVR4 participates in the epigenetic defense mechanism by introducing H3K9me3 marks to repress potentially harmful transposon activity . SUVR4 specifically converts H3K9me1 into H3K9me3 at transposons and pseudogenes within the euchromatin [93, 94], but SUVR1 and SUVR2 show no detected histone methyltransferase activity in vitro [92, 95]. In this study, D. catenatum DcSUVR4, P. equestris PeSUVR4, and rice SDG712 were grouped together with Arabidopsis SUVR4 but not with SUVR1/2, implying that they possess ubiquitin binding and HKMTase activity, except for PeSUVR4, which includes an obviously short sequence and lacks WIYLD and PreSET domains.
Clade V-5 contains one member in each tested species, and characterized by an additional tandem ZnF_C2H2 domain, except for SDG706. In Arabidopsis, SUVR5 lacks the SRA domain but recognizes specific DNA sequences through its zinc finger motifs and establishes the heterochromatic state through H3K9me2 deposition in a DNA methylation-independent manner . The knockout of SUVR5 leads to delayed flowering, and no further enhanced phenotype occurs in the quintuple suvr1 suvr2 suvr3 suvr4 suvr5 mutants [96, 97]. This finding suggests that SUVR5 is a dominant developmental regulator in SUVR subclass.
Clade V-6 members exist in one copy in each species, and their encoding proteins are notably shorter than those of the other clades in this class. Arabidopsis SUVR3 contains an additional AWS domain close to the SET–PostSET domain combination, and DcSUVR3 contains an intact PreSET–SET–PostSET domain combination. However, SUVR3 orthologs in rice and in P. equestris only contain a PreSET domain, suggesting that the genes in Clade V-6 may undergo less selective pressures and become increasingly divergent during evolution. The functions of the genes in this clade remain uncharacterized thus far.
Tissue and organ expression profiles of DcSDG genes
To investigate the potential roles of DcSDGs during growth and development in D. catenatum, we detected the expression profiles of DcSDGs by reanalyzing the RNA-seq data from different plant tissues and organs, including leaf, root, green root tip, white part of root, stem, flower bud, sepal, labellum (lip), pollinia, and gynostemium (column) .
Based on hierarchical clustering (Fig. 6 and Additional file 3), the expression patterns of DcSDGs could be divided into two groups, G1 and G2. G1 genes usually feature low expressions in most tissues and organs. However, several genes are highly expressed in specific tissues and organs, such as DcATX3d in flower bud, DcASHR3 in root and flower bud, and DcATXR5a, DcSUVH5a, and DcATX3a in pollinia. G2 genes display diverse expression profiles in diﬀerent tissues and organs. The majority of G2 group genes are highly expressed in most of the detected tissues and organs, whereas several show intermediate expressions in most tissues and organs, such as DcASHH3a, DcASHH1, DcSUVH5b, and DcSDG47. Furthermore, ~86% of genes (38/44) present intermediate and high expressions in flower buds, compared with ~77% (34) in root and gynostemium, ~57% (25) in leaf, ~45% (20) in stem, ~43% (19) in pollinia, ~55% (24) in sepal, and ~64% (28) in labellum. These findings suggest that DcSDG family plays essential roles in flower bud formation. Notably, DcSDG51 in leaf, DcSUVH2b in pollinia, and DcSUVH1a in root and stem display the most distinguished expressions, indicating their prominent functions in specific tissues.
The expression profiles of six duplicated DcSDG gene pairs were further compared (Additional file 4). In general, one copy showed higher expression levels than the others in all tissues, except for DcSUVH5a/5b pairs, suggesting that one paralog might performed a dominant function during plant growth and development. DcASHH3a/3b, DcATX3b/3c, DcATXR5a/5b and DcSUVR14b/14c exhibited similar expression patterns, whereas DcSUVH5a/5b pairs displayed differential expression profiles in the detected tissues and organs. These results indicate that distinct duplicated gene pairs might undergo different evolutionary pressures and diverge at varying periods.
Expression levels of DcSDGs in response to environmental stresses
D. catenatum is an epiphytic orchid plant that grows on trunks and cliffs and often experiences diverse environmental stresses, such as drought, cold, and high temperature. To detect the responses of DcSDG genes to drought stress, the expression profiles of DcSDGs were assessed by analyzing the RNA-seq data from the leaves under different drought treatments  (Fig. 7 and Additional file 5). In brief, the seedlings were irrigated on the 1st day, and no water was provided from the 2nd day to the 7th day, and recovered on the 8th day. Leaves were sampled at both 06:30 and 18:30 on the 2nd (DR5 and DR8), 7th (DR6 and DR10), and 9th (DR7 and DR15) days, respectively, and at 18:30 on the 8th day (DR11). The results showed that one-week of drought stress notably repressed the expressions of DcCLF, DcASHR3, DcSUVR3, and DcSUVR14c, but obviously induced the expression of DcATXR5b, DcATXR4, and DcSDG49 when sampling at both dawn and dusk. Subsequently, rewatering restored the expression levels of DcASHR3, DcSUVR3, DcATXR5b, DcATXR4, and DcSDG49.
To explore the possible roles of SDG proteins in response to cold stress, we evaluated the expression levels of DcSDGs through analyzing the raw RNA-seq reads from the leaves of D. catenatum seedlings treated at 20 ℃ (control) and 0 ℃ for 20 h, respectively  (Fig. 8 and Additional file 6). Data showed that 32% of DcSDG genes (14) showed transcription change in response to cold stress. For example, genes with upregulated expression consisted of DcASHH1 (II), DcATX3b (III), DcSUVH4 (V), DcSUVH5a (V), DcSUVH5b (V), DcSDG45 (VII), and DcSDG51 (VII), whereas genes with deregulated expressions included DcATXR5a (IV), DcATXR5b (IV), DcSUVR14a (V), DcASHR1 (VI), DcASHR2 (VI), DcATXR2 (VI), and DcSDG50 (VII). The expression levels of DcSUVH5a, DcATXR5a, DcASHR2 and DcSUVR14a (fold change >2 or <0.5) are significantly influenced by cold.
To further understand the roles of the DcSDG proteins in response to high temperature (35℃) stress, the expression profiles of SDG genes in the leaves of D. catenatum seedlings were examined by quantitative reverse-transcription–polymerase chain reaction (RT-qPCR) (Fig. 9). The results show the diverse expression patterns of DcSDG genes during heat shock treatment. At 3 h after treatment (HAT), the number of upregulated genes (10) was slightly higher than that of downregulated genes (7). At 6 HAT, more DcSDG genes were induced (15 upregulated genes versus 10 downregulated genes). At 12 HAT, the number of upregulated genes (27) was evidently higher than that of downregulated genes (3). Of the genes examined upon exposure to heat shock, three Class II genes (DcASHH3a/3b, DcATX3a), five Class V genes (DcSUVH2a/2b, DcSUR14b/14c, DcSUVR3), two Class VI genes (DcATXR1 and DcASHR1), two Class VII genes (DcSDG45/48), and one Class M gene DcATXR3 were distinguished from the corresponding control in at least at one time point (P<0.05, Fig. 9).